From August 19 to 23, the 2026 World Robot Conference (WRC) was held at Beijing Etrong International Exhibition & Convention Center. As a global leader in motion capture, CHINGMU joined forces with enterprises and industrial platforms including Unitree Technology, DexRobot, JD.com, Yizhuang Robotics, China Academy of Quality Inspection and Testing Sciences, Jijia Tech and EndecodeX. We also collaborated with government authorities, universities and industry organizations such as Beijing Economic‑Technological Development Area, Beijing Municipal Bureau of Culture and Tourism, Zhongguancun Beijing‑Tianjin‑Hebei New Energy Vehicle Collaborative Development Promotion Association, and Academy of Arts & Design, Tsinghua University. Adopting a format combining multiple zones at the main venue plus sub‑venues, seven exhibition spots showcased the complete workflow covering data acquisition, teleoperation control, robot execution and motion evaluation.

01丨Wearing Optical Motion Capture on Hands: OptiUMI Redefines Portable Data Acquisition

During the conference, CHINGMU and DexRobot officially unveiled OptiUMI, a high‑precision portable multimodal data acquisition system built for embodied intelligence training and robot manipulation learning scenarios.

Current gripper‑based data acquisition products mostly rely on inertial sensors, failing to meet training‑grade precision requirements; alternatively, they require fixed lab deployment and cannot collect data in real‑world task environments. As the world’s first commercially mature product integrating "UMI + optical motion capture", OptiUMI features head‑mounted infrared optical tracking core. It brings professional‑grade spatial positioning capability of optical motion capture into wearable hardware. The full system consists of two OptiUMI handheld grippers and one P3H head‑mounted mocap camera. It delivers sub‑millimeter 3D positioning and synchronously captures multimodal manipulation data including RGB footage and 6DoF pose as well as force‑sensing information.

• Sub‑millimeter Precision: Infrared optical tracking enables high‑accuracy 3D spatial positioning

• ≤50 μs Hardware‑level Multimodal Synchronization: Hardware‑based time alignment for multi‑channel sensor data with minimal error

• 120 FPS High‑frame‑rate 3D Tracking: Real‑time output of high‑frequency 6DoF pose data

• Real‑time 3D Reprojection Validation: 3D tracking results overlaid onto RGB video on the fly for instant data quality inspection

• Hand‑Object Synchronized Tracking: Simultaneously acquires 6DoF poses of grippers and manipulated objects to fully record spatial interaction

• Dual‑coordinate System (Ego / World): Freely switch between first‑person and world coordinates for imitation learning and scene reproduction

Boasting lightweight wearable design without complicated peripherals, the system can be rapidly deployed in real‑world settings such as laboratories, industrial production lines, warehouses and households. Human demonstration operations are converted into multimodal datasets via OptiUMI. These datasets support URDF reconstruction, imitation learning and algorithm validation on simulation platforms, and can also be applied for physical robot training and policy deployment. Human manipulation is thus converted into training‑grade data "interpretable and learnable" for robots.

02丨 Project Decode: High‑precision Spatial Perception Underpins the Vision Foundation for Vehicle‑Machine Integration

At the Vehicle‑Machine Integration Test Site, CHINGMU, EndecodeX and Zhongguancun Beijing‑Tianjin‑Hebei New Energy Vehicle Collaborative Development Promotion Association presented Project Decode, the multimodal high‑quality data acquisition system. Targeting relatively fixed operation scenarios such as workstations and production lines, the solution integrates optical, inertial, tactile sensing and video technologies. It enables one‑stop synchronous acquisition of motion, video, tactile, object and environmental data in real‑world settings, generating quantifiable and reproducible test & training datasets for cockpit interaction, driver monitoring, robot teleoperation and human‑vehicle‑machine collaborative missions.

Within CHINGMU’s full‑lifecycle robotics capabilities, Project Decode undertakes the role of "data collection". Data acquisition is no longer limited to merely "recording motions", but continuously generates data assets for robot learning amid real‑world tasks.

03丨 Teleoperation: Robots as Human "External Avatars"

        What comes next with high‑quality datasets?

        During WRC, CHINGMU, Beijing Economic‑Technological Development Area and Academy of Arts & Design, Tsinghua University co‑staged the "External Avatar" real‑time teleoperation dance show for humanoid robots. Every movement performed by the dancer was captured and solved in real‑time by CHINGMU’s high‑precision motion capture system and mapped onto the humanoid robot. The robot danced synchronously with the human performer as if it were the human’s external avatar.

The demonstration illustrates RoboTeleop, the robot teleoperation capability within CHINGMU RoboDecode Full‑life‑cycle Robotics Platform. Human operators control robot movements in real‑time, while operation logs are reversely generated into reusable datasets. Compared with offline‑only data collection, teleoperation places humans at the very front‑end of robot training pipelines: it serves both as instant motion output channel and source of continuous high‑quality data accumulation.

04丨 Human‑Robot Competition: Robot Accurately Strikes Balls under High‑speed Dynamic Conditions

        One of the most popular demonstration zones at the conference featured robot table‑tennis matches. CHINGMU, Unitree Technology and Jijia Tech delivered exciting human‑vs‑robot table‑tennis contests across different setups.

        Playing table‑tennis is far more complex for robots than it appears. 

        A table‑tennis ball can travel at over ten meters per second. Only hundreds of milliseconds elapse from serve to landing, accompanied by complex spin and unpredictable trajectories. Robots must accomplish ball detection, 3D positioning, trajectory prediction, hitting planning and motion control within extremely short time windows. The critical prerequisite is to accurately obtain real‑time spatial positions and motion trajectories of both the ball and robot itself — only with accurate perception can precise strikes be achieved.

        CHINGMU motion‑capture hardware tracks position and velocity of the high‑speed flying table‑tennis ball in real‑time, calculates target position, orientation and velocity for ball‑hitting, and transmits parameters down to underlying motion‑control systems to realize high‑speed human‑robot rallies.

        Such real‑world motion scenarios deliver values in data generation and validation. High‑precision spatial motion data enables robot movements to be recorded, analyzed and evaluated rather than merely visually completed.

        From dancing to table‑tennis, robot motions are evolving from pre‑scripted routines toward dynamic and complex real‑world environments.

05丨Quantify Every Movement: RoboEval Makes Robot Performance Visible

        What metric defines the quality of robot motions?

        At the sub‑venue of WRC, CHINGMU, Beijing Municipal Bureau of Culture and Tourism and China Academy of Quality Inspection and Testing Sciences showcased evaluation demos of RoboEval Robot Motion Evaluation Platform. By "quantifying every movement and validating every delivery", RoboEval closes the final gap within the "data‑to‑motion" workflow.

        Built upon CHINGMU’s high‑precision optical motion capture and multi‑source data fusion capabilities, RoboEval conducts quantitative analysis on robot motion trajectories, 6DoF posture stability, response speed, repeatability and execution deviation. It supports synchronized data input from force‑measuring platforms and video streams. Covering robot R&D validation, factory acceptance testing, post‑repair re‑testing and version regression, RoboEval builds a closed‑loop motion‑quality evaluation workflow spanning R&D to product delivery. Users can view data reports online in real‑time or export evaluation reports with one click. Deviations, jitter and instability can be localized and quantified to feed back into algorithm optimization.

This marks the shift of robot "motion capability" from subjective observation toward data‑driven and standardized validation. Robots are expected not only to execute actions, but also deliver measurable stability, precision and quantifiable performance gaps against target benchmarks.

The End‑to‑end "Data‑Motion‑Evaluation" Workflow for Full‑life‑cycle Robots

        From data acquisition by OptiUMI and Project Decode, real‑time teleoperation, robot execution in real‑world scenarios such as table‑tennis games, down to quantitative evaluation powered by RoboEval, CHINGMU demonstrated a full‑lifecycle technical workflow for robotics at WRC.

        This capability corresponds to the overall architecture of CHINGMU RoboDecode Full‑life‑cycle Robotics Platform. Built on optical motion capture and multimodal perception, it incorporates seven functional modules: Acquisition, Fusion, Aggregation, Mapping, Training, Control and Evaluation. It forms a complete closed‑loop covering real‑world data capture, training‑dataset processing, dataset construction, motion retargeting, embodied training & learning, teleoperation control as well as motion evaluation & quality verification. It supports continuous robot iteration across R&D tuning, mass‑production delivery, after‑sales repair and operational validation.

Hardware is merely the entry point. While the industry keeps exploring "what kind of data robots need", CHINGMU aims to build robotic data infrastructure that connects acquisition, training adaptation and evaluation. Efficient data circulation and continuous feedback constitute the true foundation for large‑scale deployment of embodied intelligence.